Exercise: Telling Stories with Data

• What does statistics mean to you?
• How about data science?
• data visualization?

Telling Stories with Data

• What is Statistics?
  • hypothesis tests
  • pattern finding
  • predictive modeling
  • storytelling with data can help you solve real-world problems (predicting unrest, decreasing crime) or it can help you stay more informed

Compelling - Hans Rosling

http://www.youtube.com/embed/jbkSRLYSojo?rel=0

Exercise: Hans Rosling Discussion

• What did you learn from this movie?
• How did Hans Rosling use data visualization to tell a story?
• What principles from the visualization would you like to be able to do?

Exercise: Why use Graphics

• Why do you, or have you, in the past used data graphics?
• What types of graphs have you used (feel free to sketch them)?

Why use Graphics

• Why do you, or have you, in the past used data graphics?
  • Exploratory Graphics
  • Publication Graphics
  • Presentation Graphics

Exercise: Visualizing Patterns Over Time

• What are we looking for with data over time?

Solution: Visualizing Patterns Over Time

• What are we looking for with data over time?
  • Trends (increasing/decreasing)
  • Are season cycles present?
• Identifying these patterns requires looking beyond single points
• We are also interested in looking at more the data in more detail
  • Are there outliers?
  • Do any time periods look out of place?
  • Are there spikes or dips?
  • What causes any of these irregularities?

Capital Bikeshare Data

library(readr)
bike.data <- read_csv('http://www.math.montana.edu/ahoegh/teaching/stat408/datasets/Bike.csv')

## Parsed with column specification:
## cols(
##   datetime = col_datetime(format = ""),
##   season = col_double(),
##   holiday = col_double(),
##   workingday = col_double(),
##   weather = col_double(),
##   temp = col_double(),
##   atemp = col_double(),
##   humidity = col_double(),
##   windspeed = col_double(),
##   casual = col_double(),
##   registered = col_double(),
##   count = col_double()
## )

Capital Bikeshare Data

library(lubridate)

## 
## Attaching package: 'lubridate'

## The following object is masked from 'package:base':
## 
##     date

library(dplyr)
bike.data <- bike.data %>% mutate(year = as.factor(year(datetime)), month = as.factor(month(datetime)))
monthly.counts <- bike.data %>% group_by(month) %>% summarize(num_bikes = sum(count)) %>% arrange(month)
monthly.counts

## # A tibble: 12 x 2
##    month num_bikes
##    <fct>     <dbl>
##  1 1         79884
##  2 2         99113
##  3 3        133501
##  4 4        167402
##  5 5        200147
##  6 6        220733
##  7 7        214617
##  8 8        213516
##  9 9        212529
## 10 10       207434
## 11 11       176440
## 12 12       160160

Discrete Points: Bar Charts

Discrete Points: Bar Charts - Code

library(ggplot2)
monthly.counts %>% 
  ggplot(aes(y = num_bikes, x = month)) + 
  geom_bar(stat = 'identity') + xlab('Month') + 
  ylab('Bike Rentals') + 
  labs(title = 'Bike Rentals per Month in 2011-2012 \n Capital Bikeshare in Washington, DC', 
       caption = 'Source: www.capitalbikeshare.com')

Discrete Points: Stacked Bar

Discrete Points in Time: Stacked Bar - Code

bike.counts <- aggregate(cbind(bike.data$casual,bike.data$registered),
                         by=list(bike.data$month), sum)
barplot(t(as.matrix(bike.counts[,-1])), 
        names.arg =collect(select(monthly.counts, month))[[1]], 
        xlab='Month', sub ='Source: www.capitalbikeshare.com', 
        ylab='Bike Rentals', 
        main='Bike Rentals per Month in 2011 - 2012 \n Capital Bikeshare in Washington, DC',
        col=c("darkblue","red"),legend.text = c("Casual", "Registered"),
        args.legend = list(x = "topleft"))

Discrete Points in Time: Points

Discrete Points in Time: Points

Discrete Points in Time: Points

plot(rowSums(bike.counts[,-1])~bike.counts[,1],xlab='Month',
     sub ='Source: www.capitalbikeshare.com', ylab='Bike Rentals', 
     main='Bike Rentals per Month \n Capital Bikeshare in Washington, DC',
     col=c("darkblue"),pch=16,axes=F,
     ylim=c(0,max(rowSums(bike.counts[,-1]))))
axis(2)
axis(1,at=1:12)
box()

Connect the Dots

Connect the Dots

mean_temp <- bike.data %>% group_by(month) %>%
  summarize(mean_temp = mean(temp)) %>% 
  mutate(month = as.numeric(month))

ggplot(aes(y=temp, x= month), data = bike.data) +
  geom_jitter(alpha = .1) + 
  geom_line(inherit.aes = F, aes(y = mean_temp, x = month),
            data = mean_temp, color = 'red', lwd = 2) +
  ylab('Average Temp (C)') + xlab('Month') + 
  labs(title = 'Average Temperature in Washington, DC', 
                       caption = 'Source: www.capitalbikeshare.com')

Exercise: Patterns over Time

Consider the number of bike rentals per hour per season

• Is this an example of continuous or discrete time?
• Make a figure to display your findings

Solution: Patterns over Time

Solution: Patterns over Time

bike.data$season <- as.character(bike.data$season)
ggplot(bike.data, aes(season,count, color = season)) 
+ geom_jitter(alpha = .01) + geom_violin() +
  stat_summary(fun.y=mean, geom="point", shape=23, size=2) 
+ labs(title='Average number of bike rentals / hour during the four seasons, where 1 = Jan - Mar', 
       caption = 'Source: Capital BikeShare') + 
  theme_dark()

Exercise: Visualizing Proportions

• What to look for in proportions?

Visualizing Proportions

• What to look for in proportions?
  • Generally looking for maximum, minimum, and overall distribution.
• Many of the figures we have discussed are useful here as well: for example, stacked bar charts or points to look at changes in proportions over time.
• Another possibility, which we will not cover, are plotting with rectangles known as a tree map.

Exercise: Visualizing Relationships

• When considering relationships between variables, what are we looking for?

Visualizing Relationships

• When considering relationships between variables, what are we looking for?
  • If something goes up, do other variables have a positive relationship, negative relationship, or no relationship.
  • What is the distribution of your data? (both univariate and multivariate)

Relationships: Scatterplots

Visualizing Relationships: Scatterplots - code

bike.data$tempF <- bike.data$temp * 1.8 + 32
plot(bike.data$count~bike.data$tempF,pch=16,
     col=rgb(100,0,0,10,max=255),ylab='Hourly Bike Rentals',
     xlab='Temp (F)',sub ='Source: www.capitalbikeshare.com',
     main='Hourly Bike Rentals by Temperature')
bike.fit <- loess(count~tempF,bike.data)
temp.seq <- seq(min(bike.data$tempF),max(bike.data$tempF))
lines(predict(bike.fit,temp.seq)~temp.seq,lwd=2)

Visualizing Relationships: Multivariate Scatterplots

pairs(bike.data[,c(12,15,8)])

Multivariate Scatterplots

Relationships: Multivariate Scatterplots

par(mfcol=c(2,2),oma = c(1,0,0,0))
bike.data$tempF <- bike.data$temp * 1.8 + 32
plot(bike.data$count~bike.data$tempF,pch=16,col=rgb(100,0,0,10,max=255),
     ylab='Hourly Bike Rentals',xlab='Temp (F)',
     main='Hourly Bike Rentals by Temperature')
bike.fit <- loess(count~tempF,bike.data)
temp.seq <- seq(min(bike.data$tempF),max(bike.data$tempF))
lines(predict(bike.fit,temp.seq)~temp.seq,lwd=2)

plot(bike.data$count~bike.data$humidity,pch=16,
     col=rgb(100,0,100,10,max=255),
     ylab='Hourly Bike Rentals',xlab='Humidity (%)',
     main='Hourly Bike Rentals by Humidity')
bike.fit <- loess(count~humidity,bike.data)
humidity.seq <- seq(min(bike.data$humidity),max(bike.data$humidity))
lines(predict(bike.fit,humidity.seq)~humidity.seq,lwd=2)

plot(bike.data$count~bike.data$windspeed,pch=16,col=rgb(0,0,100,10,max=255),
     ylab='Hourly Bike Rentals',xlab='Windspeed (MPH)',main='Hourly Bike Rentals by Windspeed')
bike.fit <- loess(count~windspeed,bike.data)
windspeed.seq <- seq(min(bike.data$windspeed),max(bike.data$windspeed))
lines(predict(bike.fit,windspeed.seq)~windspeed.seq,lwd=2)

plot(bike.data$count~as.factor(bike.data$weather),col=rgb(0,100,0,255,max=255),
     ylab='Hourly Bike Rentals',xlab='Weather Conditions',main='Hourly Bike Rentals by Weather')

mtext('Source: www.capitalbikeshare.com', outer = TRUE, cex = .9, side=1)
par(mfcol=c(1,1),oma = c(0,0,0,0))

Relationships: Histograms

Relationships: Histograms

hist(bike.data$tempF,prob=T, main='Temperature (F)',col='red',xlab='')

Multiple Histograms

Visualizing Relationships: Multiple Histograms - Code

par(mfrow=c(2,1))
bike.data$reltempF <- bike.data$atemp * 1.8 + 32

hist(bike.data$tempF,prob=T,breaks='FD',
     main='Temperature (F)',col='red',xlab='',
     xlim=c(0,max(c(bike.data$reltempF,bike.data$tempF))))
hist(bike.data$reltempF,prob=T,breaks='FD',
     main='Relative Temperature (F)',col='orange',xlab='',
     xlim=c(0,max(c(bike.data$reltempF,bike.data$tempF))))

Exercise: Visualizing Relationships

• Summarize (visually) some relationship from the bike data set